Anthropic Proposes AI Safety Bill to Limit Federal Kill-Switch Power

Anthropic is advocating for a new AI safety bill that would establish specific thresholds for blocking the distribution of AI models, as TechCrunch first reported via Politico.

The proposed legislation defines a "foreseeable, imminent risk" as a scenario where more than 50 people could die or property damage could exceed one billion dollars. These catastrophic risks must specifically result from a cyberattack, the model evading developer controls, or the creation of a biological weapon. TechCrunch notes these metrics mirror those found in California's Senate Bill 53.

Under the draft bill, the process for halting a model's release would rely on "independent evaluators" registered with the U.S. Commerce Department. These evaluators could be paid by the AI labs, provided the payment is not tied to the results of the evaluation. If an evaluator identifies an imminent risk, they have 72 hours to notify the Commerce Department.

Crucially, the legislation would strip federal agencies of the unilateral power to stop a model's distribution. Instead, Politico reports the Commerce Department would be required to petition a judge in Washington, D.C., to request a halt to the release.

A spokesperson for Anthropic confirmed the draft bill to Politico, stating the proposal aligns with the company's Advanced AI Framework, which it has promoted to state and federal lawmakers since June. This approach contrasts with other industry views, such as that of Nvidia CEO Jensen Huang, who recently argued that no new laws are necessary for the AI industry.